(2007) Journal of Cognitive Engineering and Decisi

1(2007)JournalofCognitiveEngineeringandDecisionMaking,1(2).121-147.IncludingamodelofVisualProcessingwithaCognitiveArchitecturetoModelaSimpleTeleoperationTask24February2007FrankE.Ritter2,3,UrmilaKukreja2,andRobertSt.Amant11DepartmentofComputerScience,NorthCarolinaStateUniversityRaleigh,NC27695USA2DepartmentofComputerScienceandEngineering,ThePennsylvaniaStateUniversityUniversityPark,PA16802USA3CollegeofInformationSciencesandTechnology,ThePennsylvaniaStateUniversityUniversityPark,PA16802USAAbstractThisarticlepresentsprogressinprovidingusermodelswithsufficientvisualinformationandmotorcontroltoperformteleoperationwithanunmodified,physicallyrealizedrobot.Usermodelsbuiltbyextendingcognitivemodelstointeractdirectlywithinterfacescanprovideatheoreticalbasisforpredictinguserbehavior.Thesemodelscanhelpsummarizeandexplainusabilityissuesindomainsinwhichconventionalusertestingistootime-consuming,toodemandingofotherresources,ortoodynamicforstaticmodels.TheusermodelconsistsofanACT-RcognitivemodelandtheSegManimageprocessingandinterpretationsystem.ACT-Rsupportsdirectingsimplerovernavigationandresponsetimepredictions.SegMansupportsinterpretingmanyaspectsofHCIinterfacesandcannowinterpretsimpleaspectsofvideousedinsimplenavigationtasksandprovidekeypressesandmouseactionsdirectly.Processinglimitedamountsofanimageasahumanfoveahelpedmakethissystemworkinrealtime.Astudyinrobotteleoperationprovidesevidencethatthecognitiveandperceptual/motormodelapproximateshumanbehaviour(basedoncomparisonoftasktime,learningbehavior,andmouseactions)inasimplenavigationtask.Thisworkdemonstrateshowusermodellingtechniquesarematuringtotheextentthattheycanbeusedforassessinginterfacesfordynamictasksbypredictingperformanceduringteleoperation,acommonHRItask.2IntroductionAscomputerandroboticstechnologyhasprogressed,systemshavesimultaneouslybecomemorecapable,moreaffordable,andamorecommonpartoftheworkenvironmentsandeverydaylivesofawiderangeofusers.Human-robotinteraction(HRI)andhuman-computerinteraction(HCI)shareacentralconcernofusability.Severaldevelopersofrobotsforteleoperationhavearguedthatusabilityisimportant.Theynotethatoperatingarobotisacomplextask(Casper&Murphy,2003);understandinghowoperatorsuserobotswillremainaproblemforsometime,andisofinteresttorobotdevelopers(Cradall,Nielsen,&Goodrich,2003;Drury,Hestand,Yanco,&Scholtz,2004;Marble,Bruemmer,Few,&Dudenhoeffer,2004;OlsenJr.&Goodrich,2003;Scholtz,2003;Scholtz,Antonishek,&Young,2003).Severalrobotdevelopershavegonesofarastoargueandpresentevidencethatrobotsthatarenotusablewillnotbeused(Murphy,Blitch,&Casper,2002).WhileHCIevaluationtechniquesarerobustandmayhaveclearapplicationtoHRI,HRIposesmoredifficultproblemsforevaluationbecauseitcoversawiderrangeoftasksandenvironments.HRIextensionstonewtasksandenvironmentsopenthepossibilitythatnovelapproachestoevaluationmaybeuseful.TherearealsootherdifferencesbetweenHCIandHRIthatmakesomeoftheevaluationtechniqueslessapplicable.Scholtz(2003)reviewsthesedifferences,andweexaminethemindetailinthenextsection.Yanco,Drury,andScholtz(2004)identifythreeclassesofevaluationmethodscommoninHCIthatmightbeusedinHRI:inspectionmethodscarriedoutbyinterfacedesignexperts;empiricalmethodsbasedonobservationandtestingwithusers;andformalmethodsbasedonanalyticalmodelsofinteraction.TheseclassesofevaluationapproacheshaveresultedinsignificantadvancesinourunderstandingofinteractioninbothHCIandHRI.InspectionmethodsforimprovingconventionaluserinterfaceshasbeendocumentedextensivelybyNielsen(1993);theevaluationguidelinesprovidedbyScholtz(2002)andrefinedbyYancoetal.(2004)canbeviewedasahigh-levelsetofinspectionmethodsthatprovideinsightintoHRItasks.Empiricalevaluationbasedonuserstudies,bothinthelaboratoryandinthefield,arestaplesinHCIandHRIdevelopment(e.g.,Scholtz,Antonishek,&Young,2003);descriptionsofinterfaceprojectsaregenerallyconsideredincompletewithoutsomereferencetousertesting.Formalmathematicalmethods,includingusermodels,whilelesscommon,havebeenusedtorigorouslyidentifyandanalyzeusabilityflawsineverydaysystemssuchascalculators(Thimbleby,2002).WeareonlyawareoflimitedresearchontheuseofusermodelsforHRIapplications.Thistypeofwork,ontheapplicationofGOMSLmodels3forrepresentinguserbehaviorinasimplerovernavigation(Kaber,Wang,&Kim,2006),andonmobilerobotsreplicatinghuman“hideandseek”tasks(Traftonetal.,2006),hasprimarilyfocusedonhigh-levelmodelingapproaches.Thesemethodsallinfluencedesignbothdirectlythroughfeedbackonaspecificinterfaceandindirectlybydesignersmodifyingtheirdesigninanticipationofthefeedback.ThefocusofthisarticleisonatechniqueforHRIevaluationbasedondetailedcognitivemodelling.Acomputationalcognitivemodelisatheoryofhumancognitionthatisrealizedasacomputerprogram.Cognitivemodellingprovidesawayofapplyingwhatisknownabouthumanbehaviorbothtoexplainandtoreproducehumanbehaviorinspecifictasks.Cognitivearchitecturesandmodelscreatedinthemintegratetheoriesofcognition,visualattention,andmotormovement.Theywereinitiallycreatedasscientifictheoriestounderstandcognition(e.g.,Newell&Simon,1972)buttheyincreasinglyhavebeenusedtoeval